From 2D to 3D Using Neural Nets technical online lecture
In this talk we present a new artificial intelligence implementation which takes as input a 2D image and automatically reconstructs a 3D model. The reconstruction can happen in any resolution. We see how this same architecture combined with a generative adversarial network (GAN), similar in type to the network use for deep-fake, can be used to generate new 3D models.
We discuss some of the challenges with 3D modelling and AI, we will present cool implementations of AI in visualization, texture analysis and 3D modelling.
PDF of the talk:
During the talk, I followed the implicit decoder research:
Open source code of the research (including trained network and datasets)
https://github.com/czq142857/implicit-decoder
My own two blog posts about the research:
